We assess the validity of mortality analysis based on retrospective reporting of smoking habits by subjects and proxies after a time lag of approximately ten years. The analysis is based on 2855 subject respondents and 615 proxy respondents to the NHANES I Epidemiologic Followup Survey (NHEFS), a national followup study of persons examined in the National Health and Nutrition Examination Survey (NHANES I). Persons 45-77 years of age at baseline were selected for analysis. Using three categories of smoking at baseline (current, former, never), the retrospective classification (NHEFS) matched the NHANES I classification for 89 per cent of the subject respondents and 83 per cent of the proxy respondents. Agreement levels were higher for women (92 per cent of subjects and 89 per cent of proxies) than for men (85 and 80 per cent). We used Cox regression models to assess the use of retrospective smoking information in an analysis of the relationship between selected risk factors and mortality for persons 45-64 years of age. There were few substantive differences in results with use of NHEFS smoking data versus NHANES I smoking data. In particular, the effects of age, systolic blood pressure, education, and race on mortality were not sensitive to the source of the smoking data (NHANES I or NHEFS) or the form of the smoking variable (two-, three-, or six-level categorization). The effects of body mass index on mortality were sensitive to the form of the smoking variable but results based on NHANES I and NHEFS were quite similar. These results suggest that smoking information obtained from proxy respondents is adequate for analyses of the risk of total mortality associated with smoking and other risk factors. Further assessment is needed to determine the adequacy of such data for analyses of mortality from specific causes that are sensitive to amount smoked or duration of smoking.